The post gives a more detailed description.
Tuesday, June 8, 2010
This post by the FME Evangelist describes how to configure your Workspace so that a minimum number of features are cached in memory. The less amount of caching, the more efficient your workspace. This applies to any workspace that has more than one writer (aka destination) dataset in it. Always put the dataset that has the potential to contain the most features to the top of the writer list.